Exploring the Effectiveness of Multi-stage Fine-tuning for Cross-encoder Re-rankers
Fuente:
arXiv
Salvato in:
| Autori principali: | , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2025
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866913764494802944 |
|---|---|
| author | Pezzuti, Francesca MacAvaney, Sean Tonellotto, Nicola |
| author_facet | Pezzuti, Francesca MacAvaney, Sean Tonellotto, Nicola |
| contents | State-of-the-art cross-encoders can be fine-tuned to be highly effective in passage re-ranking. The typical fine-tuning process of cross-encoders as re-rankers requires large amounts of manually labelled data, a contrastive learning objective, and a set of heuristically sampled negatives. An alternative recent approach for fine-tuning instead involves teaching the model to mimic the rankings of a highly effective large language model using a distillation objective. These fine-tuning strategies can be applied either individually, or in sequence. In this work, we systematically investigate the effectiveness of point-wise cross-encoders when fine-tuned independently in a single stage, or sequentially in two stages. Our experiments show that the effectiveness of point-wise cross-encoders fine-tuned using contrastive learning is indeed on par with that of models fine-tuned with multi-stage approaches. Code is available for reproduction at https://github.com/fpezzuti/multistage-finetuning. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2503_22672 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Exploring the Effectiveness of Multi-stage Fine-tuning for Cross-encoder Re-rankers Pezzuti, Francesca MacAvaney, Sean Tonellotto, Nicola Information Retrieval Artificial Intelligence State-of-the-art cross-encoders can be fine-tuned to be highly effective in passage re-ranking. The typical fine-tuning process of cross-encoders as re-rankers requires large amounts of manually labelled data, a contrastive learning objective, and a set of heuristically sampled negatives. An alternative recent approach for fine-tuning instead involves teaching the model to mimic the rankings of a highly effective large language model using a distillation objective. These fine-tuning strategies can be applied either individually, or in sequence. In this work, we systematically investigate the effectiveness of point-wise cross-encoders when fine-tuned independently in a single stage, or sequentially in two stages. Our experiments show that the effectiveness of point-wise cross-encoders fine-tuned using contrastive learning is indeed on par with that of models fine-tuned with multi-stage approaches. Code is available for reproduction at https://github.com/fpezzuti/multistage-finetuning. |
| title | Exploring the Effectiveness of Multi-stage Fine-tuning for Cross-encoder Re-rankers |
| topic | Information Retrieval Artificial Intelligence |
| url | https://arxiv.org/abs/2503.22672 |